First professional review of Intel's Rocket Lake (11700K) is out.. and it's a disaster

[Delicieuxz]

Can someone fill me in on why this CPU even exists if it performs worse than the model it's supposed to be replacing, while featuring the same cores/threads and having a higher peak power consumption?

[Video Beagle]

Is this the CPU that was in this mess?

 

[D13H4RD]

17 minutes ago, Video Beagle said:

Is this the CPU that was in this mess?

 

Nope. That was Tiger Lake. 

[Jet_ski]

1 hour ago, D13H4RD said:

Nope. That was Tiger Lake. 

I’m sure Intel is working hard on the next nonsensical presentation

[Monkey Dust]

I guess a bios or microcode update might release a bit more performance, but I can't imagine a huge uplift?

[Rym]

Huh I guess AMD really is ryzen.

 

Get it? 

[porina]

12 hours ago, illegalwater said:

Performance gains range from mediocre to non-existent, and in the worst cases it's actually slower than Skylake, overall it's still slower vs the 5800X despite consuming more than 2x the power.

For the power it is easy to take it out of context if you don't compare like with like. If you look at the AVX2 power, it is same ball park as Comet Lake and Coffee Lake. This is within expectations and not surprising.

 

AVX-512 power is elevated, but that implementation can give a significant performance uplift. You use more power, you do a LOT more work, for software that supports it. That's the double edge sword Intel has. While it was already introduced in HEDT, mobile, and server, it has been lacking on desktop until this generation. Software that uses it limited, but that which can, gets a significant performance boost. For now, continuing on the thinking that most software doesn't use it, the CPU also wont reach those higher power levels.

 

This is mostly a perception fault anyway. These high peak values are allowed by Intel without voiding CPU warranty, as opposed to AMD's restriction in that area. So not surprising many higher end mobo default to practically unlimited power. It would be interesting to see an actual power limited run, since in most cases the perf loss would be minor compared to the power reduction.

 

9 hours ago, FakeKGB said:

This reminds me of FX.

Lots of heat for not an amazing a lot of performance.

FX basically sucked so hard at doing anything I'd rather have an Intel 4 core of that era than an 8 core FX. What we have here is not anywhere near that level of difference.

 

5 hours ago, sora.sky said:

Their insistence on bringing over big.LITTLE has me even more concerned with stability and scheduling issues on the OS side. I hope they make SKUs with only big cores or the ability to disable the little cores. 

Rumours/leaks appear to show models with no "little" cores.

 

2 hours ago, Delicieuxz said:

Can someone fill me in on why this CPU even exists if it performs worse than the model it's supposed to be replacing, while featuring the same cores/threads and power TDP?

Short answer is, it isn't, while also bringing with it new features such as PCIe 4.0 support and AVX-512 for anyone that might care about either.

[heickelrrx]

8 core 16 thread with Xe IGPU

Basically best gaming CPU period, since dedicated card do not exist in 2021

[WolframaticAlpha]

12 minutes ago, heickelrrx said:

8 core 16 thread with Xe IGPU

Basically best gaming CPU period, since dedicated card do not exist in 2021

You can get an older card(1060/1050ti) and bundle in a Ryzen 3/5 basedmin Zen 3 which will give better perf in gaming

[WolframaticAlpha]

Seriously, I hope intel really hammers AMD with alder lake. I can't honestly see them getting beat up more. This bloodbath by AMD is too gory. 

 

Lisa Su, stop. Have mercy on them.

[ne0tic]

Okay, I'm keeping my scuffed 5900X and B550 board for a bit longer lol. Hopefully Alder Lake is good or AMD can fix all the USB issues and BIOS glitches! 

[Empire24453]

So, point saying that I use a i7 10700K myself and yet all in all it's still far better then the new upcoming 11700K,  I mean I get get 5.2ghz easy and yet 11700K does 5ghz on paper and base clock is even less at 3.6ghz. I mean they going down in performance not up. Yes PCI at 4 and not 3, yet big deal. 

 

No point upgraded at all, rather stick to 10th gen rather then 11th, I'm expecting 10900K to be far better then 11900K by a long gang of new york. 11700K is 5% lower then 10700K for GTA for flip sacks on 1440P at high settings  Cost point - OMG, it's just no point. 

[leadeater]

1 hour ago, porina said:

AVX-512 power is elevated, but that implementation can give a significant performance uplift. You use more power, you do a LOT more work, for software that supports it. That's the double edge sword Intel has

Even then that could be a slight understatement. Using AVX-512 it's roughly twice the power and 6 times the performance, perf per watt for AVX-512 is really damn good, even if total power is high. But it's a desktop performance CPU, let it be 500W for all I care.

[CarlBar]

4 hours ago, Delicieuxz said:

Can someone fill me in on why this CPU even exists if it performs worse than the model it's supposed to be replacing, while featuring the same cores/threads and power TDP?

 

Short answer, it doesn't perform worse than the model it's replacing the vast majority of the time, it just doesn't have a huge lead either, it's a marginal improvement that comes almost entirely from IPC.

[Castdeath97]

*Sigh*, at least it might help CPU supplies.

[cj09beira]

52 minutes ago, WolframaticAlpha said:

Seriously, I hope intel really hammers AMD with alder lake. I can't honestly see them getting beat up more. This bloodbath by AMD is too gory. 

 

Lisa Su, stop. Have mercy on them.

with shortages it might not matter but amd is still too small for comfort at the speed its going its going to take a while for amd to get enough market share to be in a comfortable place 

[porina]

Just now, leadeater said:

Even then that could be a slight understatement. Using AVX-512 it's roughly twice the power and 6 times the performance, perf per watt for AVX-512 is really damn good, even if total power is high. But it's a desktop performance CPU, let it be 500W for all I care.

There is one test result in that review that really doesn't make sense to me, and it is the one you're looking at. That perf increase seems too much but we don't have enough context into understanding it. I'd expect ball park 2x over AVX2 CPUs for FP64 workloads. So either it is using something other than FP64, or the AVX2 implementation is less optimal. Anandtech's x86 CPU test policy is to run fixed sets of benchmarks, settings which can and will be argued over. Understanding what AVX-512 is really doing here will require a deeper look than that can give. I'll await actual pricing before deciding if I'll actually get one to play with. In a similar way I was hyped to bench Zen 3 until I saw the pricing and checked out.

 

Looking at y-cruncher the test there was done at 250m digits, which would use around 1.25 GB of ram for data. So we can disregard the multithread result for microarchitecture analysis as it is simply ram performance limited. The code is somewhat varied to my understanding, so it wont be hitting the same compute units as you might with Prime95 for example. In single thread where ram would be much less of a limitation, we do see something closer to 2x (actually less) ball park improvement from previous Intel.

[leadeater]

32 minutes ago, porina said:

There is one test result in that review that really doesn't make sense to me, and it is the one you're looking at. That perf increase seems too much but we don't have enough context into understanding it. I'd expect ball park 2x over AVX2 CPUs for FP64 workloads. So either it is using something other than FP64, or the AVX2 implementation is less optimal.

I think it might be more a case of AVX-512 being a bit more optimized for the cache layout than AVX2 can benefit from. I'm thinking even though the AVX-512 operations are larger they still fit within the cache, with a higher data utilizations of it, where as with AVX2 while it also fits within the cache too it's not able to do more operations per cycle either due to limitations in the cache size not allowing more or the decode and dispatch not allowing more, it's already at maximum per cycle yet not completely filling the data and op caches.

 

Would need to checkout the wikichip architecture page when it's out and figure out if it could be something like the above.

 

So I think it's a case of AVX2 performance being limited by the front end.

[Moonzy]

56 minutes ago, leadeater said:

let it be 500W for all I care

no, that would mean i would have to upgrade my PSU... AGAIN

 

2016: yea boi, i got a 620W psu, should be good for years

nvidia launching 30-series: lul cute

intel rocket: haha psu go brrr

[leadeater]

1 minute ago, Moonzy said:

no, that would mean i would have to upgrade my PSU... AGAIN

 

2016: yea boi, i got a 620W psu, should be good for years

nvidia launching 30-series: lul cute

intel rocket: haha psu go brrr

 

Nah m8, noooo problem 

[Thaldor]

So, Rocket Lake i9's run hot, what's the news here?

.

.

.

It's just i7... Just the i7-x700K... Not the i9, just the i7...

 

 

[Delicieuxz]

1 hour ago, CarlBar said:

Short answer, it doesn't perform worse than the model it's replacing the vast majority of the time, it just doesn't have a huge lead either, it's a marginal improvement that comes almost entirely from IPC.

In AnadTech's non-gaming testing, it's better. But for the large majority of AnandTech's gaming benchmarks, the 11700k delivers a lower average FPS than the 10700k. Together with it having a notably higher peak power consumption than the 10700k, I think it's a dud and I'd probably rather have the 10700k.

[Prodigy_Smit]

The wording for the title lead me to believe that the review was bad. 

Oh no no no, this is much worse, FX series reincarnated under a different company. 

[heickelrrx]

Just now, Prodigy_Smit said:

The wording for the title lead me to believe that the review was bad. 

Oh no no no, this is much worse, FX series reincarnated under a different company. 

Nope

Rocket Lake is Very usable and practical CPU, 

FX is just... Not Practical to use, Run Extremely Hot, Get Destroyed by mere 4 core CPU
No one match FX Disasters

[LAwLz]

This thread just reminded me that the average LTT user doesn't understand benchmarking or technology in general...

 

AVX512 vs AVX2 for power consumption numbers...

GPU bottlenecked benchmarks for CPU results... 

 

I know that most users here sees the world in black and white, and right now that world is "Intel bad. AMD good". But come on... At least be a bit objective here. 

 

 

Rocket Lake is not great. In a lot of ways its really bad. There is no need to exaggerate or misconstrue the findings of the article to make it look bad. 

 

Unless the rocket lake chips are significantly cheaper than AMD's 5000 series, or you really need AVX512, then there isn't really any point in getting this. Which is kind of what we expected to begin with. Rocket lake is a stopgap solution before alder lake. 

Personally I'd recommend not buying anything right now. Even if pricing and availability was decent, we are so close to a new generation of four example DDR that if you wait 2 years or so, you will get something much better for the same money. More so than your typical "wait 2 years and get something better".